Facebook's AI Pluribus Beats Top Poker Pros in 6-Player Game

Facebook's AI program, Pluribus, has achieved a significant milestone by defeating a group of professional poker players in a 6-max no-limit Texas hold'em (NLHE) game. The professional players included Linus Løliger, Nick Petrangelo, Darren Elias, Greg Merson, and others.

Pluribus was developed by Noam Brown and Tuomas Sandholm, researchers from Carnegie Mellon University. The experiments were conducted on two 14-core processors with 128 GB of memory, not on a supercomputer. Pluribus learned to play poker by playing against itself for eight days on a server with a 64-core processor.

The details of the algorithm used by Pluribus will not be published to avoid harming the online poker industry. In the first experiment, five Pluribus instances played against one professional player, and in the second experiment, five professional players played against one Pluribus instance. The experiments involved a total of 10,000 hands, with blinds set at 50/100 and a stack of 10,000 chips for each hand. In the second experiment, Pluribus won 4.77 big blinds per 100 hands with a standard deviation of 2.5.